net/netfilter/nfnetlink_acct.c

   1 /*
   2  * (C) 2011 Pablo Neira Ayuso <pablo@netfilter.org>
   3  * (C) 2011 Intra2net AG <http://www.intra2net.com>
   4  *
   5  * This program is free software; you can redistribute it and/or modify
   6  * it under the terms of the GNU General Public License version 2 as
   7  * published by the Free Software Foundation (or any later at your option).
   8  */
   9 #include <linux/init.h>
  10 #include <linux/module.h>
  11 #include <linux/kernel.h>
  12 #include <linux/skbuff.h>
  13 #include <linux/atomic.h>
  14 #include <linux/netlink.h>
  15 #include <linux/rculist.h>
  16 #include <linux/slab.h>
  17 #include <linux/types.h>
  18 #include <linux/errno.h>
  19 #include <net/netlink.h>
  20 #include <net/sock.h>
  21
  22 #include <linux/netfilter.h>
  23 #include <linux/netfilter/nfnetlink.h>
  24 #include <linux/netfilter/nfnetlink_acct.h>
  25
  26 MODULE_LICENSE("GPL");
  27 MODULE_AUTHOR("Pablo Neira Ayuso <pablo@netfilter.org>");
  28 MODULE_DESCRIPTION("nfacct: Extended Netfilter accounting infrastructure");
  29
  30 struct nf_acct {
  31         atomic64_t              pkts;
  32         atomic64_t              bytes;
  33         unsigned long           flags;
  34         struct list_head        head;
  35         atomic_t                refcnt;
  36         char                    name[NFACCT_NAME_MAX];
  37         struct rcu_head         rcu_head;
  38         char                    data[0];
  39 };
  40
  41 struct nfacct_filter {
  42         u32 value;
  43         u32 mask;
  44 };
  45
  46 #define NFACCT_F_QUOTA (NFACCT_F_QUOTA_PKTS | NFACCT_F_QUOTA_BYTES)
  47 #define NFACCT_OVERQUOTA_BIT    2       /* NFACCT_F_OVERQUOTA */
  48
  49 static int nfnl_acct_new(struct net *net, struct sock *nfnl,
  50                          struct sk_buff *skb, const struct nlmsghdr *nlh,
  51                          const struct nlattr * const tb[])
  52 {
  53         struct nf_acct *nfacct, *matching = NULL;
  54         char *acct_name;
  55         unsigned int size = 0;
  56         u32 flags = 0;
  57
  58         if (!tb[NFACCT_NAME])
  59                 return -EINVAL;
  60
  61         acct_name = nla_data(tb[NFACCT_NAME]);
  62         if (strlen(acct_name) == 0)
  63                 return -EINVAL;
  64
  65         list_for_each_entry(nfacct, &net->nfnl_acct_list, head) {
  66                 if (strncmp(nfacct->name, acct_name, NFACCT_NAME_MAX) != 0)
  67                         continue;
  68
  69                 if (nlh->nlmsg_flags & NLM_F_EXCL)
  70                         return -EEXIST;
  71
  72                 matching = nfacct;
  73                 break;
  74         }
  75
  76         if (matching) {
  77                 if (nlh->nlmsg_flags & NLM_F_REPLACE) {
  78                         /* reset counters if you request a replacement. */
  79                         atomic64_set(&matching->pkts, 0);
  80                         atomic64_set(&matching->bytes, 0);
  81                         smp_mb__before_atomic();
  82                         /* reset overquota flag if quota is enabled. */
  83                         if ((matching->flags & NFACCT_F_QUOTA))
  84                                 clear_bit(NFACCT_OVERQUOTA_BIT,
  85                                           &matching->flags);
  86                         return 0;
  87                 }
  88                 return -EBUSY;
  89         }
  90
  91         if (tb[NFACCT_FLAGS]) {
  92                 flags = ntohl(nla_get_be32(tb[NFACCT_FLAGS]));
  93                 if (flags & ~NFACCT_F_QUOTA)
  94                         return -EOPNOTSUPP;
  95                 if ((flags & NFACCT_F_QUOTA) == NFACCT_F_QUOTA)
  96                         return -EINVAL;
  97                 if (flags & NFACCT_F_OVERQUOTA)
  98                         return -EINVAL;
  99                 if ((flags & NFACCT_F_QUOTA) && !tb[NFACCT_QUOTA])
 100                         return -EINVAL;
 101
 102                 size += sizeof(u64);
 103         }
 104
 105         nfacct = kzalloc(sizeof(struct nf_acct) + size, GFP_KERNEL);
 106         if (nfacct == NULL)
 107                 return -ENOMEM;
 108
 109         if (flags & NFACCT_F_QUOTA) {
 110                 u64 *quota = (u64 *)nfacct->data;
 111
 112                 *quota = be64_to_cpu(nla_get_be64(tb[NFACCT_QUOTA]));
 113                 nfacct->flags = flags;
 114         }
 115
 116         strncpy(nfacct->name, nla_data(tb[NFACCT_NAME]), NFACCT_NAME_MAX);
 117
 118         if (tb[NFACCT_BYTES]) {
 119                 atomic64_set(&nfacct->bytes,
 120                              be64_to_cpu(nla_get_be64(tb[NFACCT_BYTES])));
 121         }
 122         if (tb[NFACCT_PKTS]) {
 123                 atomic64_set(&nfacct->pkts,
 124                              be64_to_cpu(nla_get_be64(tb[NFACCT_PKTS])));
 125         }
 126         atomic_set(&nfacct->refcnt, 1);
 127         list_add_tail_rcu(&nfacct->head, &net->nfnl_acct_list);
 128         return 0;
 129 }
 130
 131 static int
 132 nfnl_acct_fill_info(struct sk_buff *skb, u32 portid, u32 seq, u32 type,
 133                    int event, struct nf_acct *acct)
 134 {
 135         struct nlmsghdr *nlh;
 136         struct nfgenmsg *nfmsg;
 137         unsigned int flags = portid ? NLM_F_MULTI : 0;
 138         u64 pkts, bytes;
 139         u32 old_flags;
 140
 141         event |= NFNL_SUBSYS_ACCT << 8;
 142         nlh = nlmsg_put(skb, portid, seq, event, sizeof(*nfmsg), flags);
 143         if (nlh == NULL)
 144                 goto nlmsg_failure;
 145
 146         nfmsg = nlmsg_data(nlh);
 147         nfmsg->nfgen_family = AF_UNSPEC;
 148         nfmsg->version = NFNETLINK_V0;
 149         nfmsg->res_id = 0;
 150
 151         if (nla_put_string(skb, NFACCT_NAME, acct->name))
 152                 goto nla_put_failure;
 153
 154         old_flags = acct->flags;
 155         if (type == NFNL_MSG_ACCT_GET_CTRZERO) {
 156                 pkts = atomic64_xchg(&acct->pkts, 0);
 157                 bytes = atomic64_xchg(&acct->bytes, 0);
 158                 smp_mb__before_atomic();
 159                 if (acct->flags & NFACCT_F_QUOTA)
 160                         clear_bit(NFACCT_OVERQUOTA_BIT, &acct->flags);
 161         } else {
 162                 pkts = atomic64_read(&acct->pkts);
 163                 bytes = atomic64_read(&acct->bytes);
 164         }
 165         if (nla_put_be64(skb, NFACCT_PKTS, cpu_to_be64(pkts)) ||
 166             nla_put_be64(skb, NFACCT_BYTES, cpu_to_be64(bytes)) ||
 167             nla_put_be32(skb, NFACCT_USE, htonl(atomic_read(&acct->refcnt))))
 168                 goto nla_put_failure;
 169         if (acct->flags & NFACCT_F_QUOTA) {
 170                 u64 *quota = (u64 *)acct->data;
 171
 172                 if (nla_put_be32(skb, NFACCT_FLAGS, htonl(old_flags)) ||
 173                     nla_put_be64(skb, NFACCT_QUOTA, cpu_to_be64(*quota)))
 174                         goto nla_put_failure;
 175         }
 176         nlmsg_end(skb, nlh);
 177         return skb->len;
 178
 179 nlmsg_failure:
 180 nla_put_failure:
 181         nlmsg_cancel(skb, nlh);
 182         return -1;
 183 }
 184
 185 static int
 186 nfnl_acct_dump(struct sk_buff *skb, struct netlink_callback *cb)
 187 {
 188         struct net *net = sock_net(skb->sk);
 189         struct nf_acct *cur, *last;
 190         const struct nfacct_filter *filter = cb->data;
 191
 192         if (cb->args[2])
 193                 return 0;
 194
 195         last = (struct nf_acct *)cb->args[1];
 196         if (cb->args[1])
 197                 cb->args[1] = 0;
 198
 199         rcu_read_lock();
 200         list_for_each_entry_rcu(cur, &net->nfnl_acct_list, head) {
 201                 if (last) {
 202                         if (cur != last)
 203                                 continue;
 204
 205                         last = NULL;
 206                 }
 207
 208                 if (filter && (cur->flags & filter->mask) != filter->value)
 209                         continue;
 210
 211                 if (nfnl_acct_fill_info(skb, NETLINK_CB(cb->skb).portid,
 212                                        cb->nlh->nlmsg_seq,
 213                                        NFNL_MSG_TYPE(cb->nlh->nlmsg_type),
 214                                        NFNL_MSG_ACCT_NEW, cur) < 0) {
 215                         cb->args[1] = (unsigned long)cur;
 216                         break;
 217                 }
 218         }
 219         if (!cb->args[1])
 220                 cb->args[2] = 1;
 221         rcu_read_unlock();
 222         return skb->len;
 223 }
 224
 225 static int nfnl_acct_done(struct netlink_callback *cb)
 226 {
 227         kfree(cb->data);
 228         return 0;
 229 }
 230
 231 static const struct nla_policy filter_policy[NFACCT_FILTER_MAX + 1] = {
 232         [NFACCT_FILTER_MASK]    = { .type = NLA_U32 },
 233         [NFACCT_FILTER_VALUE]   = { .type = NLA_U32 },
 234 };
 235
 236 static struct nfacct_filter *
 237 nfacct_filter_alloc(const struct nlattr * const attr)
 238 {
 239         struct nfacct_filter *filter;
 240         struct nlattr *tb[NFACCT_FILTER_MAX + 1];
 241         int err;
 242
 243         err = nla_parse_nested(tb, NFACCT_FILTER_MAX, attr, filter_policy);
 244         if (err < 0)
 245                 return ERR_PTR(err);
 246
 247         if (!tb[NFACCT_FILTER_MASK] || !tb[NFACCT_FILTER_VALUE])
 248                 return ERR_PTR(-EINVAL);
 249
 250         filter = kzalloc(sizeof(struct nfacct_filter), GFP_KERNEL);
 251         if (!filter)
 252                 return ERR_PTR(-ENOMEM);
 253
 254         filter->mask = ntohl(nla_get_be32(tb[NFACCT_FILTER_MASK]));
 255         filter->value = ntohl(nla_get_be32(tb[NFACCT_FILTER_VALUE]));
 256
 257         return filter;
 258 }
 259
 260 static int nfnl_acct_get(struct net *net, struct sock *nfnl,
 261                          struct sk_buff *skb, const struct nlmsghdr *nlh,
 262                          const struct nlattr * const tb[])
 263 {
 264         int ret = -ENOENT;
 265         struct nf_acct *cur;
 266         char *acct_name;
 267
 268         if (nlh->nlmsg_flags & NLM_F_DUMP) {
 269                 struct netlink_dump_control c = {
 270                         .dump = nfnl_acct_dump,
 271                         .done = nfnl_acct_done,
 272                 };
 273
 274                 if (tb[NFACCT_FILTER]) {
 275                         struct nfacct_filter *filter;
 276
 277                         filter = nfacct_filter_alloc(tb[NFACCT_FILTER]);
 278                         if (IS_ERR(filter))
 279                                 return PTR_ERR(filter);
 280
 281                         c.data = filter;
 282                 }
 283                 return netlink_dump_start(nfnl, skb, nlh, &c);
 284         }
 285
 286         if (!tb[NFACCT_NAME])
 287                 return -EINVAL;
 288         acct_name = nla_data(tb[NFACCT_NAME]);
 289
 290         list_for_each_entry(cur, &net->nfnl_acct_list, head) {
 291                 struct sk_buff *skb2;
 292
 293                 if (strncmp(cur->name, acct_name, NFACCT_NAME_MAX)!= 0)
 294                         continue;
 295
 296                 skb2 = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
 297                 if (skb2 == NULL) {
 298                         ret = -ENOMEM;
 299                         break;
 300                 }
 301
 302                 ret = nfnl_acct_fill_info(skb2, NETLINK_CB(skb).portid,
 303                                          nlh->nlmsg_seq,
 304                                          NFNL_MSG_TYPE(nlh->nlmsg_type),
 305                                          NFNL_MSG_ACCT_NEW, cur);
 306                 if (ret <= 0) {
 307                         kfree_skb(skb2);
 308                         break;
 309                 }
 310                 ret = netlink_unicast(nfnl, skb2, NETLINK_CB(skb).portid,
 311                                         MSG_DONTWAIT);
 312                 if (ret > 0)
 313                         ret = 0;
 314
 315                 /* this avoids a loop in nfnetlink. */
 316                 return ret == -EAGAIN ? -ENOBUFS : ret;
 317         }
 318         return ret;
 319 }
 320
 321 /* try to delete object, fail if it is still in use. */
 322 static int nfnl_acct_try_del(struct nf_acct *cur)
 323 {
 324         int ret = 0;
 325
 326         /* we want to avoid races with nfnl_acct_find_get. */
 327         if (atomic_dec_and_test(&cur->refcnt)) {
 328                 /* We are protected by nfnl mutex. */
 329                 list_del_rcu(&cur->head);
 330                 kfree_rcu(cur, rcu_head);
 331         } else {
 332                 /* still in use, restore reference counter. */
 333                 atomic_inc(&cur->refcnt);
 334                 ret = -EBUSY;
 335         }
 336         return ret;
 337 }
 338
 339 static int nfnl_acct_del(struct net *net, struct sock *nfnl,
 340                          struct sk_buff *skb, const struct nlmsghdr *nlh,
 341                          const struct nlattr * const tb[])
 342 {
 343         char *acct_name;
 344         struct nf_acct *cur;
 345         int ret = -ENOENT;
 346
 347         if (!tb[NFACCT_NAME]) {
 348                 list_for_each_entry(cur, &net->nfnl_acct_list, head)
 349                         nfnl_acct_try_del(cur);
 350
 351                 return 0;
 352         }
 353         acct_name = nla_data(tb[NFACCT_NAME]);
 354
 355         list_for_each_entry(cur, &net->nfnl_acct_list, head) {
 356                 if (strncmp(cur->name, acct_name, NFACCT_NAME_MAX) != 0)
 357                         continue;
 358
 359                 ret = nfnl_acct_try_del(cur);
 360                 if (ret < 0)
 361                         return ret;
 362
 363                 break;
 364         }
 365         return ret;
 366 }
 367
 368 static const struct nla_policy nfnl_acct_policy[NFACCT_MAX+1] = {
 369         [NFACCT_NAME] = { .type = NLA_NUL_STRING, .len = NFACCT_NAME_MAX-1 },
 370         [NFACCT_BYTES] = { .type = NLA_U64 },
 371         [NFACCT_PKTS] = { .type = NLA_U64 },
 372         [NFACCT_FLAGS] = { .type = NLA_U32 },
 373         [NFACCT_QUOTA] = { .type = NLA_U64 },
 374         [NFACCT_FILTER] = {.type = NLA_NESTED },
 375 };
 376
 377 static const struct nfnl_callback nfnl_acct_cb[NFNL_MSG_ACCT_MAX] = {
 378         [NFNL_MSG_ACCT_NEW]             = { .call = nfnl_acct_new,
 379                                             .attr_count = NFACCT_MAX,
 380                                             .policy = nfnl_acct_policy },
 381         [NFNL_MSG_ACCT_GET]             = { .call = nfnl_acct_get,
 382                                             .attr_count = NFACCT_MAX,
 383                                             .policy = nfnl_acct_policy },
 384         [NFNL_MSG_ACCT_GET_CTRZERO]     = { .call = nfnl_acct_get,
 385                                             .attr_count = NFACCT_MAX,
 386                                             .policy = nfnl_acct_policy },
 387         [NFNL_MSG_ACCT_DEL]             = { .call = nfnl_acct_del,
 388                                             .attr_count = NFACCT_MAX,
 389                                             .policy = nfnl_acct_policy },
 390 };
 391
 392 static const struct nfnetlink_subsystem nfnl_acct_subsys = {
 393         .name                           = "acct",
 394         .subsys_id                      = NFNL_SUBSYS_ACCT,
 395         .cb_count                       = NFNL_MSG_ACCT_MAX,
 396         .cb                             = nfnl_acct_cb,
 397 };
 398
 399 MODULE_ALIAS_NFNL_SUBSYS(NFNL_SUBSYS_ACCT);
 400
 401 struct nf_acct *nfnl_acct_find_get(struct net *net, const char *acct_name)
 402 {
 403         struct nf_acct *cur, *acct = NULL;
 404
 405         rcu_read_lock();
 406         list_for_each_entry_rcu(cur, &net->nfnl_acct_list, head) {
 407                 if (strncmp(cur->name, acct_name, NFACCT_NAME_MAX)!= 0)
 408                         continue;
 409
 410                 if (!try_module_get(THIS_MODULE))
 411                         goto err;
 412
 413                 if (!atomic_inc_not_zero(&cur->refcnt)) {
 414                         module_put(THIS_MODULE);
 415                         goto err;
 416                 }
 417
 418                 acct = cur;
 419                 break;
 420         }
 421 err:
 422         rcu_read_unlock();
 423         return acct;
 424 }
 425 EXPORT_SYMBOL_GPL(nfnl_acct_find_get);
 426
 427 void nfnl_acct_put(struct nf_acct *acct)
 428 {
 429         if (atomic_dec_and_test(&acct->refcnt))
 430                 kfree_rcu(acct, rcu_head);
 431
 432         module_put(THIS_MODULE);
 433 }
 434 EXPORT_SYMBOL_GPL(nfnl_acct_put);
 435
 436 void nfnl_acct_update(const struct sk_buff *skb, struct nf_acct *nfacct)
 437 {
 438         atomic64_inc(&nfacct->pkts);
 439         atomic64_add(skb->len, &nfacct->bytes);
 440 }
 441 EXPORT_SYMBOL_GPL(nfnl_acct_update);
 442
 443 static void nfnl_overquota_report(struct nf_acct *nfacct)
 444 {
 445         int ret;
 446         struct sk_buff *skb;
 447
 448         skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC);
 449         if (skb == NULL)
 450                 return;
 451
 452         ret = nfnl_acct_fill_info(skb, 0, 0, NFNL_MSG_ACCT_OVERQUOTA, 0,
 453                                   nfacct);
 454         if (ret <= 0) {
 455                 kfree_skb(skb);
 456                 return;
 457         }
 458         netlink_broadcast(init_net.nfnl, skb, 0, NFNLGRP_ACCT_QUOTA,
 459                           GFP_ATOMIC);
 460 }
 461
 462 int nfnl_acct_overquota(const struct sk_buff *skb, struct nf_acct *nfacct)
 463 {
 464         u64 now;
 465         u64 *quota;
 466         int ret = NFACCT_UNDERQUOTA;
 467
 468         /* no place here if we don't have a quota */
 469         if (!(nfacct->flags & NFACCT_F_QUOTA))
 470                 return NFACCT_NO_QUOTA;
 471
 472         quota = (u64 *)nfacct->data;
 473         now = (nfacct->flags & NFACCT_F_QUOTA_PKTS) ?
 474                atomic64_read(&nfacct->pkts) : atomic64_read(&nfacct->bytes);
 475
 476         ret = now > *quota;
 477
 478         if (now >= *quota &&
 479             !test_and_set_bit(NFACCT_OVERQUOTA_BIT, &nfacct->flags)) {
 480                 nfnl_overquota_report(nfacct);
 481         }
 482
 483         return ret;
 484 }
 485 EXPORT_SYMBOL_GPL(nfnl_acct_overquota);
 486
 487 static int __net_init nfnl_acct_net_init(struct net *net)
 488 {
 489         INIT_LIST_HEAD(&net->nfnl_acct_list);
 490
 491         return 0;
 492 }
 493
 494 static void __net_exit nfnl_acct_net_exit(struct net *net)
 495 {
 496         struct nf_acct *cur, *tmp;
 497
 498         list_for_each_entry_safe(cur, tmp, &net->nfnl_acct_list, head) {
 499                 list_del_rcu(&cur->head);
 500
 501                 if (atomic_dec_and_test(&cur->refcnt))
 502                         kfree_rcu(cur, rcu_head);
 503         }
 504 }
 505
 506 static struct pernet_operations nfnl_acct_ops = {
 507         .init   = nfnl_acct_net_init,
 508         .exit   = nfnl_acct_net_exit,
 509 };
 510
 511 static int __init nfnl_acct_init(void)
 512 {
 513         int ret;
 514
 515         ret = register_pernet_subsys(&nfnl_acct_ops);
 516         if (ret < 0) {
 517                 pr_err("nfnl_acct_init: failed to register pernet ops\n");
 518                 goto err_out;
 519         }
 520
 521         pr_info("nfnl_acct: registering with nfnetlink.\n");
 522         ret = nfnetlink_subsys_register(&nfnl_acct_subsys);
 523         if (ret < 0) {
 524                 pr_err("nfnl_acct_init: cannot register with nfnetlink.\n");
 525                 goto cleanup_pernet;
 526         }
 527         return 0;
 528
 529 cleanup_pernet:
 530         unregister_pernet_subsys(&nfnl_acct_ops);
 531 err_out:
 532         return ret;
 533 }
 534
 535 static void __exit nfnl_acct_exit(void)
 536 {
 537         pr_info("nfnl_acct: unregistering from nfnetlink.\n");
 538         nfnetlink_subsys_unregister(&nfnl_acct_subsys);
 539         unregister_pernet_subsys(&nfnl_acct_ops);
 540 }
 541
 542 module_init(nfnl_acct_init);
 543 module_exit(nfnl_acct_exit);